An Automatic Actors to Threads Mapping Technique for JVM-Based Actor Frameworks

Proceedings of the 14th International Conference on Modularity

2015

Self Cite

Modular reasoning about concurrent programs is complicated by the possibility of interferences happening between any two instructions of a task (pervasive interference), and these interferences not giving out any information about the behaviors of potentially interfering concurrent tasks (oblivious interference). Reasoning about a concurrent program would be easier if a programmer modularly and statically (1) knows precisely the program points at which interferences may happen (sparse interference), and (2) has some insights into behaviors of potentially interfering tasks at these points (cognizant interference). In this work we present Panini, a core concurrent calculus which guarantees sparse interference, by controlling sharing among concurrent tasks, and cognizant interference, by controlling dynamic name bindings and accessibility of states of tasks. Panini promotes capsule-oriented programming whose concurrently running capsules own their states, communicate by asynchronous invocations of their procedures and dynamically transfer ownership. Panini limits sharing among two capsules to other capsules and futures, limits accessibility of a capsule states to only through its procedures and dispatches a procedure invocation on the static type of its receiver capsule. We formalize Panini, present its semantics and illustrate how its interference model, using behavioral contracts, enables Hoare-style modular reasoning about concurrent programs with interference.

Section: Discussionmentioning

confidence: 99%

Panini: a concurrent programming model for solving pervasive and oblivious interference

Bagherzadeh

Proceedings of the 14th International Conference on Modularity

2015

Self Cite

“…Performance optimization is one of the leading reasons for breaking abstraction boundaries (at least for actors). We have evidence to suggest that properties of capsules enables modular analysis to determine mapping from capsules to threads [27]. • Capsule-oriented specification and design.…”

Section: Ongoing Explorationsmentioning

confidence: 99%

Capsule-Oriented Programming

2015 IEEE/ACM 37th IEEE International Conference on Software Engineering

2015

Self Cite

Explicit concurrency should be abolished from all higher-level programming languages (i.e. everything exceptperhaps-plain machine code.)." Dijkstra [1] (paraphrased). A promising class of concurrency abstractions replaces explicit concurrency mechanisms with a single linguistic mechanism that combines state and control and uses asynchronous messages for communications, e.g. active objects or actors, but that doesn't remove the hurdle of understanding non-local control transfer. What if the programming model enabled programmers to simply do what they do best, that is, to describe a system in terms of its modular structure and write sequential code to implement the operations of those modules and handles details of concurrency? In a recently sponsored NSF project we are developing such a model that we call capsule-oriented programming and its realization in the Panini project. This model favors modularity over explicit concurrency, encourages concurrency correctness by construction, and exploits modular structure of programs to expose implicit concurrency.Abstract-"Explicit concurrency should be abolished from all higher-level programming languages (i.e. everything exceptperhaps-plain machine code.)." Dijkstra [1] (paraphrased). A promising class of concurrency abstractions replaces explicit concurrency mechanisms with a single linguistic mechanism that combines state and control and uses asynchronous messages for communications, e.g. active objects or actors, but that doesn't remove the hurdle of understanding non-local control transfer.What if the programming model enabled programmers to simply do what they do best, that is, to describe a system in terms of its modular structure and write sequential code to implement the operations of those modules and handles details of concurrency? In a recently sponsored NSF project we are developing such a model that we call capsule-oriented programming and its realization in the Panini project. This model favors modularity over explicit concurrency, encourages concurrency correctness by construction, and exploits modular structure of programs to expose implicit concurrency.

“…To that end, this work makes several contributions 2 : 2 We build upon our preliminary AGERE 2014 workshop paper [36]. The cVector representation and analysis techniques in this work is novel, especially the local treatment of incoming and outgoing message patterns.…”

Section: Introductionmentioning

confidence: 99%

Effectively mapping linguistic abstractions for message-passing concurrency to threads on the Java virtual machine

Upadhyaya

Proceedings of the 2015 ACM SIGPLAN International Conference on Object-Oriented Programming, Systems, Languages, and Applicatio

2015

Self Cite

Efficient mapping of message passing concurrency (MPC) abstractions to Java Virtual Machine ( JVM) threads is critical for performance, scalability, and CPU utilization; but tedious and time consuming to perform manually. In general, this mapping cannot be found in polynomial time, but we show that by exploiting the local characteristics of MPC abstractions and their communication patterns this mapping can be determined effectively. We describe our MPC abstraction to thread mapping technique, its realization in two frameworks (Panini and Akka), and its rigorous evaluation using several benchmarks from representative MPC frameworks. We also compare our technique against four default mapping techniques: thread-all, roundrobin-task-all, random-task-all and work-stealing. Our evaluation shows that our mapping technique can improve the performance by 30%-60% over default mapping techniques. These improvements are due to a number of challenges addressed by our technique namely: i) balancing the computations across JVM threads, ii) reducing the communication overheads, iii) utilizing information about cache locality, and iv) mapping MPC abstractions to threads in a way that reduces the contention between JVM threads. AbstractEfficient mapping of message passing concurrency (MPC) abstractions to Java Virtual Machine (JVM) threads is critical for performance, scalability, and CPU utilization; but tedious and time consuming to perform manually. In general, this mapping cannot be found in polynomial time, but we show that by exploiting the local characteristics of MPC abstractions and their communication patterns this mapping can be determined effectively. We describe our MPC abstraction to thread mapping technique, its realization in two frameworks (Panini and Akka), and its rigorous evaluation using several benchmarks from representative MPC frameworks. We also compare our technique against four default mapping techniques: thread-all, round-robin-task-all, random-task-all and work-stealing. Our evaluation shows that our mapping technique can improve the performance by 30%-60% over default mapping techniques. These improvements are due to a number of challenges addressed by our technique namely: i) balancing the computations across JVM threads, ii) reducing the communication overheads, iii) utilizing information about cache locality, and iv) mapping MPC abstractions to threads in a way that reduces the contention between JVM threads.